4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/freezer.h>
34 #include <linux/cgroup.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
48 #include <asm/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/pgtable.h>
51 #include <asm/mmu_context.h>
53 static void exit_mm(struct task_struct
* tsk
);
55 static inline int task_detached(struct task_struct
*p
)
57 return p
->exit_signal
== -1;
60 static void __unhash_process(struct task_struct
*p
)
63 detach_pid(p
, PIDTYPE_PID
);
64 if (thread_group_leader(p
)) {
65 detach_pid(p
, PIDTYPE_PGID
);
66 detach_pid(p
, PIDTYPE_SID
);
68 list_del_rcu(&p
->tasks
);
69 __get_cpu_var(process_counts
)--;
71 list_del_rcu(&p
->thread_group
);
76 * This function expects the tasklist_lock write-locked.
78 static void __exit_signal(struct task_struct
*tsk
)
80 struct signal_struct
*sig
= tsk
->signal
;
81 struct sighand_struct
*sighand
;
84 BUG_ON(!atomic_read(&sig
->count
));
87 sighand
= rcu_dereference(tsk
->sighand
);
88 spin_lock(&sighand
->siglock
);
90 posix_cpu_timers_exit(tsk
);
91 if (atomic_dec_and_test(&sig
->count
))
92 posix_cpu_timers_exit_group(tsk
);
95 * If there is any task waiting for the group exit
98 if (sig
->group_exit_task
&& atomic_read(&sig
->count
) == sig
->notify_count
)
99 wake_up_process(sig
->group_exit_task
);
101 if (tsk
== sig
->curr_target
)
102 sig
->curr_target
= next_thread(tsk
);
104 * Accumulate here the counters for all threads but the
105 * group leader as they die, so they can be added into
106 * the process-wide totals when those are taken.
107 * The group leader stays around as a zombie as long
108 * as there are other threads. When it gets reaped,
109 * the exit.c code will add its counts into these totals.
110 * We won't ever get here for the group leader, since it
111 * will have been the last reference on the signal_struct.
113 sig
->utime
= cputime_add(sig
->utime
, tsk
->utime
);
114 sig
->stime
= cputime_add(sig
->stime
, tsk
->stime
);
115 sig
->gtime
= cputime_add(sig
->gtime
, tsk
->gtime
);
116 sig
->min_flt
+= tsk
->min_flt
;
117 sig
->maj_flt
+= tsk
->maj_flt
;
118 sig
->nvcsw
+= tsk
->nvcsw
;
119 sig
->nivcsw
+= tsk
->nivcsw
;
120 sig
->inblock
+= task_io_get_inblock(tsk
);
121 sig
->oublock
+= task_io_get_oublock(tsk
);
122 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
123 sig
= NULL
; /* Marker for below. */
126 __unhash_process(tsk
);
130 spin_unlock(&sighand
->siglock
);
133 __cleanup_sighand(sighand
);
134 clear_tsk_thread_flag(tsk
,TIF_SIGPENDING
);
135 flush_sigqueue(&tsk
->pending
);
137 flush_sigqueue(&sig
->shared_pending
);
138 taskstats_tgid_free(sig
);
139 __cleanup_signal(sig
);
143 static void delayed_put_task_struct(struct rcu_head
*rhp
)
145 put_task_struct(container_of(rhp
, struct task_struct
, rcu
));
148 void release_task(struct task_struct
* p
)
150 struct task_struct
*leader
;
153 atomic_dec(&p
->user
->processes
);
155 write_lock_irq(&tasklist_lock
);
157 BUG_ON(!list_empty(&p
->ptrace_list
) || !list_empty(&p
->ptrace_children
));
161 * If we are the last non-leader member of the thread
162 * group, and the leader is zombie, then notify the
163 * group leader's parent process. (if it wants notification.)
166 leader
= p
->group_leader
;
167 if (leader
!= p
&& thread_group_empty(leader
) && leader
->exit_state
== EXIT_ZOMBIE
) {
168 BUG_ON(task_detached(leader
));
169 do_notify_parent(leader
, leader
->exit_signal
);
171 * If we were the last child thread and the leader has
172 * exited already, and the leader's parent ignores SIGCHLD,
173 * then we are the one who should release the leader.
175 * do_notify_parent() will have marked it self-reaping in
178 zap_leader
= task_detached(leader
);
181 write_unlock_irq(&tasklist_lock
);
183 call_rcu(&p
->rcu
, delayed_put_task_struct
);
186 if (unlikely(zap_leader
))
191 * This checks not only the pgrp, but falls back on the pid if no
192 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
195 * The caller must hold rcu lock or the tasklist lock.
197 struct pid
*session_of_pgrp(struct pid
*pgrp
)
199 struct task_struct
*p
;
200 struct pid
*sid
= NULL
;
202 p
= pid_task(pgrp
, PIDTYPE_PGID
);
204 p
= pid_task(pgrp
, PIDTYPE_PID
);
206 sid
= task_session(p
);
212 * Determine if a process group is "orphaned", according to the POSIX
213 * definition in 2.2.2.52. Orphaned process groups are not to be affected
214 * by terminal-generated stop signals. Newly orphaned process groups are
215 * to receive a SIGHUP and a SIGCONT.
217 * "I ask you, have you ever known what it is to be an orphan?"
219 static int will_become_orphaned_pgrp(struct pid
*pgrp
, struct task_struct
*ignored_task
)
221 struct task_struct
*p
;
223 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
224 if ((p
== ignored_task
) ||
225 (p
->exit_state
&& thread_group_empty(p
)) ||
226 is_global_init(p
->real_parent
))
229 if (task_pgrp(p
->real_parent
) != pgrp
&&
230 task_session(p
->real_parent
) == task_session(p
))
232 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
237 int is_current_pgrp_orphaned(void)
241 read_lock(&tasklist_lock
);
242 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
243 read_unlock(&tasklist_lock
);
248 static int has_stopped_jobs(struct pid
*pgrp
)
251 struct task_struct
*p
;
253 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
254 if (!task_is_stopped(p
))
258 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
263 * Check to see if any process groups have become orphaned as
264 * a result of our exiting, and if they have any stopped jobs,
265 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
268 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
270 struct pid
*pgrp
= task_pgrp(tsk
);
271 struct task_struct
*ignored_task
= tsk
;
274 /* exit: our father is in a different pgrp than
275 * we are and we were the only connection outside.
277 parent
= tsk
->real_parent
;
279 /* reparent: our child is in a different pgrp than
280 * we are, and it was the only connection outside.
284 if (task_pgrp(parent
) != pgrp
&&
285 task_session(parent
) == task_session(tsk
) &&
286 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
287 has_stopped_jobs(pgrp
)) {
288 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
289 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
294 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
296 * If a kernel thread is launched as a result of a system call, or if
297 * it ever exits, it should generally reparent itself to kthreadd so it
298 * isn't in the way of other processes and is correctly cleaned up on exit.
300 * The various task state such as scheduling policy and priority may have
301 * been inherited from a user process, so we reset them to sane values here.
303 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
305 static void reparent_to_kthreadd(void)
307 write_lock_irq(&tasklist_lock
);
309 ptrace_unlink(current
);
310 /* Reparent to init */
311 remove_parent(current
);
312 current
->real_parent
= current
->parent
= kthreadd_task
;
315 /* Set the exit signal to SIGCHLD so we signal init on exit */
316 current
->exit_signal
= SIGCHLD
;
318 if (task_nice(current
) < 0)
319 set_user_nice(current
, 0);
323 security_task_reparent_to_init(current
);
324 memcpy(current
->signal
->rlim
, init_task
.signal
->rlim
,
325 sizeof(current
->signal
->rlim
));
326 atomic_inc(&(INIT_USER
->__count
));
327 write_unlock_irq(&tasklist_lock
);
328 switch_uid(INIT_USER
);
331 void __set_special_pids(struct pid
*pid
)
333 struct task_struct
*curr
= current
->group_leader
;
334 pid_t nr
= pid_nr(pid
);
336 if (task_session(curr
) != pid
) {
337 detach_pid(curr
, PIDTYPE_SID
);
338 attach_pid(curr
, PIDTYPE_SID
, pid
);
339 set_task_session(curr
, nr
);
341 if (task_pgrp(curr
) != pid
) {
342 detach_pid(curr
, PIDTYPE_PGID
);
343 attach_pid(curr
, PIDTYPE_PGID
, pid
);
344 set_task_pgrp(curr
, nr
);
348 static void set_special_pids(struct pid
*pid
)
350 write_lock_irq(&tasklist_lock
);
351 __set_special_pids(pid
);
352 write_unlock_irq(&tasklist_lock
);
356 * Let kernel threads use this to say that they
357 * allow a certain signal (since daemonize() will
358 * have disabled all of them by default).
360 int allow_signal(int sig
)
362 if (!valid_signal(sig
) || sig
< 1)
365 spin_lock_irq(¤t
->sighand
->siglock
);
366 sigdelset(¤t
->blocked
, sig
);
368 /* Kernel threads handle their own signals.
369 Let the signal code know it'll be handled, so
370 that they don't get converted to SIGKILL or
371 just silently dropped */
372 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= (void __user
*)2;
375 spin_unlock_irq(¤t
->sighand
->siglock
);
379 EXPORT_SYMBOL(allow_signal
);
381 int disallow_signal(int sig
)
383 if (!valid_signal(sig
) || sig
< 1)
386 spin_lock_irq(¤t
->sighand
->siglock
);
387 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= SIG_IGN
;
389 spin_unlock_irq(¤t
->sighand
->siglock
);
393 EXPORT_SYMBOL(disallow_signal
);
396 * Put all the gunge required to become a kernel thread without
397 * attached user resources in one place where it belongs.
400 void daemonize(const char *name
, ...)
403 struct fs_struct
*fs
;
406 va_start(args
, name
);
407 vsnprintf(current
->comm
, sizeof(current
->comm
), name
, args
);
411 * If we were started as result of loading a module, close all of the
412 * user space pages. We don't need them, and if we didn't close them
413 * they would be locked into memory.
417 * We don't want to have TIF_FREEZE set if the system-wide hibernation
418 * or suspend transition begins right now.
420 current
->flags
|= PF_NOFREEZE
;
422 if (current
->nsproxy
!= &init_nsproxy
) {
423 get_nsproxy(&init_nsproxy
);
424 switch_task_namespaces(current
, &init_nsproxy
);
426 set_special_pids(&init_struct_pid
);
427 proc_clear_tty(current
);
429 /* Block and flush all signals */
430 sigfillset(&blocked
);
431 sigprocmask(SIG_BLOCK
, &blocked
, NULL
);
432 flush_signals(current
);
434 /* Become as one with the init task */
436 exit_fs(current
); /* current->fs->count--; */
439 atomic_inc(&fs
->count
);
442 current
->files
= init_task
.files
;
443 atomic_inc(¤t
->files
->count
);
445 reparent_to_kthreadd();
448 EXPORT_SYMBOL(daemonize
);
450 static void close_files(struct files_struct
* files
)
458 * It is safe to dereference the fd table without RCU or
459 * ->file_lock because this is the last reference to the
462 fdt
= files_fdtable(files
);
466 if (i
>= fdt
->max_fds
)
468 set
= fdt
->open_fds
->fds_bits
[j
++];
471 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
473 filp_close(file
, files
);
483 struct files_struct
*get_files_struct(struct task_struct
*task
)
485 struct files_struct
*files
;
490 atomic_inc(&files
->count
);
496 void put_files_struct(struct files_struct
*files
)
500 if (atomic_dec_and_test(&files
->count
)) {
503 * Free the fd and fdset arrays if we expanded them.
504 * If the fdtable was embedded, pass files for freeing
505 * at the end of the RCU grace period. Otherwise,
506 * you can free files immediately.
508 fdt
= files_fdtable(files
);
509 if (fdt
!= &files
->fdtab
)
510 kmem_cache_free(files_cachep
, files
);
515 void reset_files_struct(struct files_struct
*files
)
517 struct task_struct
*tsk
= current
;
518 struct files_struct
*old
;
524 put_files_struct(old
);
527 void exit_files(struct task_struct
*tsk
)
529 struct files_struct
* files
= tsk
->files
;
535 put_files_struct(files
);
539 void put_fs_struct(struct fs_struct
*fs
)
541 /* No need to hold fs->lock if we are killing it */
542 if (atomic_dec_and_test(&fs
->count
)) {
545 if (fs
->altroot
.dentry
)
546 path_put(&fs
->altroot
);
547 kmem_cache_free(fs_cachep
, fs
);
551 void exit_fs(struct task_struct
*tsk
)
553 struct fs_struct
* fs
= tsk
->fs
;
563 EXPORT_SYMBOL_GPL(exit_fs
);
565 #ifdef CONFIG_MM_OWNER
567 * Task p is exiting and it owned mm, lets find a new owner for it
570 mm_need_new_owner(struct mm_struct
*mm
, struct task_struct
*p
)
573 * If there are other users of the mm and the owner (us) is exiting
574 * we need to find a new owner to take on the responsibility.
578 if (atomic_read(&mm
->mm_users
) <= 1)
585 void mm_update_next_owner(struct mm_struct
*mm
)
587 struct task_struct
*c
, *g
, *p
= current
;
590 if (!mm_need_new_owner(mm
, p
))
593 read_lock(&tasklist_lock
);
595 * Search in the children
597 list_for_each_entry(c
, &p
->children
, sibling
) {
599 goto assign_new_owner
;
603 * Search in the siblings
605 list_for_each_entry(c
, &p
->parent
->children
, sibling
) {
607 goto assign_new_owner
;
611 * Search through everything else. We should not get
614 do_each_thread(g
, c
) {
616 goto assign_new_owner
;
617 } while_each_thread(g
, c
);
619 read_unlock(&tasklist_lock
);
626 * The task_lock protects c->mm from changing.
627 * We always want mm->owner->mm == mm
631 * Delay read_unlock() till we have the task_lock()
632 * to ensure that c does not slip away underneath us
634 read_unlock(&tasklist_lock
);
640 cgroup_mm_owner_callbacks(mm
->owner
, c
);
645 #endif /* CONFIG_MM_OWNER */
648 * Turn us into a lazy TLB process if we
651 static void exit_mm(struct task_struct
* tsk
)
653 struct mm_struct
*mm
= tsk
->mm
;
659 * Serialize with any possible pending coredump.
660 * We must hold mmap_sem around checking core_waiters
661 * and clearing tsk->mm. The core-inducing thread
662 * will increment core_waiters for each thread in the
663 * group with ->mm != NULL.
665 down_read(&mm
->mmap_sem
);
666 if (mm
->core_waiters
) {
667 up_read(&mm
->mmap_sem
);
668 down_write(&mm
->mmap_sem
);
669 if (!--mm
->core_waiters
)
670 complete(mm
->core_startup_done
);
671 up_write(&mm
->mmap_sem
);
673 wait_for_completion(&mm
->core_done
);
674 down_read(&mm
->mmap_sem
);
676 atomic_inc(&mm
->mm_count
);
677 BUG_ON(mm
!= tsk
->active_mm
);
678 /* more a memory barrier than a real lock */
681 up_read(&mm
->mmap_sem
);
682 enter_lazy_tlb(mm
, current
);
683 /* We don't want this task to be frozen prematurely */
684 clear_freeze_flag(tsk
);
686 mm_update_next_owner(mm
);
691 reparent_thread(struct task_struct
*p
, struct task_struct
*father
, int traced
)
693 if (p
->pdeath_signal
)
694 /* We already hold the tasklist_lock here. */
695 group_send_sig_info(p
->pdeath_signal
, SEND_SIG_NOINFO
, p
);
697 /* Move the child from its dying parent to the new one. */
698 if (unlikely(traced
)) {
699 /* Preserve ptrace links if someone else is tracing this child. */
700 list_del_init(&p
->ptrace_list
);
701 if (ptrace_reparented(p
))
702 list_add(&p
->ptrace_list
, &p
->real_parent
->ptrace_children
);
704 /* If this child is being traced, then we're the one tracing it
705 * anyway, so let go of it.
709 p
->parent
= p
->real_parent
;
712 if (task_is_traced(p
)) {
714 * If it was at a trace stop, turn it into
715 * a normal stop since it's no longer being
722 /* If this is a threaded reparent there is no need to
723 * notify anyone anything has happened.
725 if (same_thread_group(p
->real_parent
, father
))
728 /* We don't want people slaying init. */
729 if (!task_detached(p
))
730 p
->exit_signal
= SIGCHLD
;
732 /* If we'd notified the old parent about this child's death,
733 * also notify the new parent.
735 if (!traced
&& p
->exit_state
== EXIT_ZOMBIE
&&
736 !task_detached(p
) && thread_group_empty(p
))
737 do_notify_parent(p
, p
->exit_signal
);
739 kill_orphaned_pgrp(p
, father
);
743 * When we die, we re-parent all our children.
744 * Try to give them to another thread in our thread
745 * group, and if no such member exists, give it to
746 * the child reaper process (ie "init") in our pid
749 static void forget_original_parent(struct task_struct
*father
)
751 struct task_struct
*p
, *n
, *reaper
= father
;
752 struct list_head ptrace_dead
;
754 INIT_LIST_HEAD(&ptrace_dead
);
756 write_lock_irq(&tasklist_lock
);
759 reaper
= next_thread(reaper
);
760 if (reaper
== father
) {
761 reaper
= task_child_reaper(father
);
764 } while (reaper
->flags
& PF_EXITING
);
767 * There are only two places where our children can be:
769 * - in our child list
770 * - in our ptraced child list
772 * Search them and reparent children.
774 list_for_each_entry_safe(p
, n
, &father
->children
, sibling
) {
779 /* if father isn't the real parent, then ptrace must be enabled */
780 BUG_ON(father
!= p
->real_parent
&& !ptrace
);
782 if (father
== p
->real_parent
) {
783 /* reparent with a reaper, real father it's us */
784 p
->real_parent
= reaper
;
785 reparent_thread(p
, father
, 0);
787 /* reparent ptraced task to its real parent */
789 if (p
->exit_state
== EXIT_ZOMBIE
&& !task_detached(p
) &&
790 thread_group_empty(p
))
791 do_notify_parent(p
, p
->exit_signal
);
795 * if the ptraced child is a detached zombie we must collect
796 * it before we exit, or it will remain zombie forever since
797 * we prevented it from self-reap itself while it was being
798 * traced by us, to be able to see it in wait4.
800 if (unlikely(ptrace
&& p
->exit_state
== EXIT_ZOMBIE
&& task_detached(p
)))
801 list_add(&p
->ptrace_list
, &ptrace_dead
);
804 list_for_each_entry_safe(p
, n
, &father
->ptrace_children
, ptrace_list
) {
805 p
->real_parent
= reaper
;
806 reparent_thread(p
, father
, 1);
809 write_unlock_irq(&tasklist_lock
);
810 BUG_ON(!list_empty(&father
->children
));
811 BUG_ON(!list_empty(&father
->ptrace_children
));
813 list_for_each_entry_safe(p
, n
, &ptrace_dead
, ptrace_list
) {
814 list_del_init(&p
->ptrace_list
);
821 * Send signals to all our closest relatives so that they know
822 * to properly mourn us..
824 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
829 * This does two things:
831 * A. Make init inherit all the child processes
832 * B. Check to see if any process groups have become orphaned
833 * as a result of our exiting, and if they have any stopped
834 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
836 forget_original_parent(tsk
);
837 exit_task_namespaces(tsk
);
839 write_lock_irq(&tasklist_lock
);
841 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
843 /* Let father know we died
845 * Thread signals are configurable, but you aren't going to use
846 * that to send signals to arbitary processes.
847 * That stops right now.
849 * If the parent exec id doesn't match the exec id we saved
850 * when we started then we know the parent has changed security
853 * If our self_exec id doesn't match our parent_exec_id then
854 * we have changed execution domain as these two values started
855 * the same after a fork.
857 if (tsk
->exit_signal
!= SIGCHLD
&& !task_detached(tsk
) &&
858 (tsk
->parent_exec_id
!= tsk
->real_parent
->self_exec_id
||
859 tsk
->self_exec_id
!= tsk
->parent_exec_id
) &&
861 tsk
->exit_signal
= SIGCHLD
;
863 /* If something other than our normal parent is ptracing us, then
864 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
865 * only has special meaning to our real parent.
867 if (!task_detached(tsk
) && thread_group_empty(tsk
)) {
868 int signal
= ptrace_reparented(tsk
) ?
869 SIGCHLD
: tsk
->exit_signal
;
870 do_notify_parent(tsk
, signal
);
871 } else if (tsk
->ptrace
) {
872 do_notify_parent(tsk
, SIGCHLD
);
876 if (task_detached(tsk
) && likely(!tsk
->ptrace
))
878 tsk
->exit_state
= state
;
880 /* mt-exec, de_thread() is waiting for us */
881 if (thread_group_leader(tsk
) &&
882 tsk
->signal
->notify_count
< 0 &&
883 tsk
->signal
->group_exit_task
)
884 wake_up_process(tsk
->signal
->group_exit_task
);
886 write_unlock_irq(&tasklist_lock
);
888 /* If the process is dead, release it - nobody will wait for it */
889 if (state
== EXIT_DEAD
)
893 #ifdef CONFIG_DEBUG_STACK_USAGE
894 static void check_stack_usage(void)
896 static DEFINE_SPINLOCK(low_water_lock
);
897 static int lowest_to_date
= THREAD_SIZE
;
898 unsigned long *n
= end_of_stack(current
);
903 free
= (unsigned long)n
- (unsigned long)end_of_stack(current
);
905 if (free
>= lowest_to_date
)
908 spin_lock(&low_water_lock
);
909 if (free
< lowest_to_date
) {
910 printk(KERN_WARNING
"%s used greatest stack depth: %lu bytes "
912 current
->comm
, free
);
913 lowest_to_date
= free
;
915 spin_unlock(&low_water_lock
);
918 static inline void check_stack_usage(void) {}
921 static inline void exit_child_reaper(struct task_struct
*tsk
)
923 if (likely(tsk
->group_leader
!= task_child_reaper(tsk
)))
926 if (tsk
->nsproxy
->pid_ns
== &init_pid_ns
)
927 panic("Attempted to kill init!");
930 * @tsk is the last thread in the 'cgroup-init' and is exiting.
931 * Terminate all remaining processes in the namespace and reap them
932 * before exiting @tsk.
934 * Note that @tsk (last thread of cgroup-init) may not necessarily
935 * be the child-reaper (i.e main thread of cgroup-init) of the
936 * namespace i.e the child_reaper may have already exited.
938 * Even after a child_reaper exits, we let it inherit orphaned children,
939 * because, pid_ns->child_reaper remains valid as long as there is
940 * at least one living sub-thread in the cgroup init.
942 * This living sub-thread of the cgroup-init will be notified when
943 * a child inherited by the 'child-reaper' exits (do_notify_parent()
944 * uses __group_send_sig_info()). Further, when reaping child processes,
945 * do_wait() iterates over children of all living sub threads.
947 * i.e even though 'child_reaper' thread is listed as the parent of the
948 * orphaned children, any living sub-thread in the cgroup-init can
949 * perform the role of the child_reaper.
951 zap_pid_ns_processes(tsk
->nsproxy
->pid_ns
);
954 NORET_TYPE
void do_exit(long code
)
956 struct task_struct
*tsk
= current
;
959 profile_task_exit(tsk
);
961 WARN_ON(atomic_read(&tsk
->fs_excl
));
963 if (unlikely(in_interrupt()))
964 panic("Aiee, killing interrupt handler!");
965 if (unlikely(!tsk
->pid
))
966 panic("Attempted to kill the idle task!");
968 if (unlikely(current
->ptrace
& PT_TRACE_EXIT
)) {
969 current
->ptrace_message
= code
;
970 ptrace_notify((PTRACE_EVENT_EXIT
<< 8) | SIGTRAP
);
974 * We're taking recursive faults here in do_exit. Safest is to just
975 * leave this task alone and wait for reboot.
977 if (unlikely(tsk
->flags
& PF_EXITING
)) {
979 "Fixing recursive fault but reboot is needed!\n");
981 * We can do this unlocked here. The futex code uses
982 * this flag just to verify whether the pi state
983 * cleanup has been done or not. In the worst case it
984 * loops once more. We pretend that the cleanup was
985 * done as there is no way to return. Either the
986 * OWNER_DIED bit is set by now or we push the blocked
987 * task into the wait for ever nirwana as well.
989 tsk
->flags
|= PF_EXITPIDONE
;
992 set_current_state(TASK_UNINTERRUPTIBLE
);
996 exit_signals(tsk
); /* sets PF_EXITING */
998 * tsk->flags are checked in the futex code to protect against
999 * an exiting task cleaning up the robust pi futexes.
1002 spin_unlock_wait(&tsk
->pi_lock
);
1004 if (unlikely(in_atomic()))
1005 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
1006 current
->comm
, task_pid_nr(current
),
1009 acct_update_integrals(tsk
);
1011 update_hiwater_rss(tsk
->mm
);
1012 update_hiwater_vm(tsk
->mm
);
1014 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
1016 exit_child_reaper(tsk
);
1017 hrtimer_cancel(&tsk
->signal
->real_timer
);
1018 exit_itimers(tsk
->signal
);
1020 acct_collect(code
, group_dead
);
1022 if (unlikely(tsk
->robust_list
))
1023 exit_robust_list(tsk
);
1024 #ifdef CONFIG_COMPAT
1025 if (unlikely(tsk
->compat_robust_list
))
1026 compat_exit_robust_list(tsk
);
1031 if (unlikely(tsk
->audit_context
))
1034 tsk
->exit_code
= code
;
1035 taskstats_exit(tsk
, group_dead
);
1044 check_stack_usage();
1046 cgroup_exit(tsk
, 1);
1049 if (group_dead
&& tsk
->signal
->leader
)
1050 disassociate_ctty(1);
1052 module_put(task_thread_info(tsk
)->exec_domain
->module
);
1054 module_put(tsk
->binfmt
->module
);
1056 proc_exit_connector(tsk
);
1057 exit_notify(tsk
, group_dead
);
1059 mpol_put(tsk
->mempolicy
);
1060 tsk
->mempolicy
= NULL
;
1064 * This must happen late, after the PID is not
1067 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
1068 exit_pi_state_list(tsk
);
1069 if (unlikely(current
->pi_state_cache
))
1070 kfree(current
->pi_state_cache
);
1073 * Make sure we are holding no locks:
1075 debug_check_no_locks_held(tsk
);
1077 * We can do this unlocked here. The futex code uses this flag
1078 * just to verify whether the pi state cleanup has been done
1079 * or not. In the worst case it loops once more.
1081 tsk
->flags
|= PF_EXITPIDONE
;
1083 if (tsk
->io_context
)
1086 if (tsk
->splice_pipe
)
1087 __free_pipe_info(tsk
->splice_pipe
);
1090 /* causes final put_task_struct in finish_task_switch(). */
1091 tsk
->state
= TASK_DEAD
;
1095 /* Avoid "noreturn function does return". */
1097 cpu_relax(); /* For when BUG is null */
1100 EXPORT_SYMBOL_GPL(do_exit
);
1102 NORET_TYPE
void complete_and_exit(struct completion
*comp
, long code
)
1110 EXPORT_SYMBOL(complete_and_exit
);
1112 asmlinkage
long sys_exit(int error_code
)
1114 do_exit((error_code
&0xff)<<8);
1118 * Take down every thread in the group. This is called by fatal signals
1119 * as well as by sys_exit_group (below).
1122 do_group_exit(int exit_code
)
1124 struct signal_struct
*sig
= current
->signal
;
1126 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
1128 if (signal_group_exit(sig
))
1129 exit_code
= sig
->group_exit_code
;
1130 else if (!thread_group_empty(current
)) {
1131 struct sighand_struct
*const sighand
= current
->sighand
;
1132 spin_lock_irq(&sighand
->siglock
);
1133 if (signal_group_exit(sig
))
1134 /* Another thread got here before we took the lock. */
1135 exit_code
= sig
->group_exit_code
;
1137 sig
->group_exit_code
= exit_code
;
1138 sig
->flags
= SIGNAL_GROUP_EXIT
;
1139 zap_other_threads(current
);
1141 spin_unlock_irq(&sighand
->siglock
);
1149 * this kills every thread in the thread group. Note that any externally
1150 * wait4()-ing process will get the correct exit code - even if this
1151 * thread is not the thread group leader.
1153 asmlinkage
void sys_exit_group(int error_code
)
1155 do_group_exit((error_code
& 0xff) << 8);
1158 static struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
1160 struct pid
*pid
= NULL
;
1161 if (type
== PIDTYPE_PID
)
1162 pid
= task
->pids
[type
].pid
;
1163 else if (type
< PIDTYPE_MAX
)
1164 pid
= task
->group_leader
->pids
[type
].pid
;
1168 static int eligible_child(enum pid_type type
, struct pid
*pid
, int options
,
1169 struct task_struct
*p
)
1173 if (type
< PIDTYPE_MAX
) {
1174 if (task_pid_type(p
, type
) != pid
)
1179 * Do not consider detached threads that are
1182 if (task_detached(p
) && !p
->ptrace
)
1185 /* Wait for all children (clone and not) if __WALL is set;
1186 * otherwise, wait for clone children *only* if __WCLONE is
1187 * set; otherwise, wait for non-clone children *only*. (Note:
1188 * A "clone" child here is one that reports to its parent
1189 * using a signal other than SIGCHLD.) */
1190 if (((p
->exit_signal
!= SIGCHLD
) ^ ((options
& __WCLONE
) != 0))
1191 && !(options
& __WALL
))
1194 err
= security_task_wait(p
);
1198 if (type
!= PIDTYPE_PID
)
1200 /* This child was explicitly requested, abort */
1201 read_unlock(&tasklist_lock
);
1205 static int wait_noreap_copyout(struct task_struct
*p
, pid_t pid
, uid_t uid
,
1206 int why
, int status
,
1207 struct siginfo __user
*infop
,
1208 struct rusage __user
*rusagep
)
1210 int retval
= rusagep
? getrusage(p
, RUSAGE_BOTH
, rusagep
) : 0;
1214 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1216 retval
= put_user(0, &infop
->si_errno
);
1218 retval
= put_user((short)why
, &infop
->si_code
);
1220 retval
= put_user(pid
, &infop
->si_pid
);
1222 retval
= put_user(uid
, &infop
->si_uid
);
1224 retval
= put_user(status
, &infop
->si_status
);
1231 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1232 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1233 * the lock and this task is uninteresting. If we return nonzero, we have
1234 * released the lock and the system call should return.
1236 static int wait_task_zombie(struct task_struct
*p
, int noreap
,
1237 struct siginfo __user
*infop
,
1238 int __user
*stat_addr
, struct rusage __user
*ru
)
1240 unsigned long state
;
1241 int retval
, status
, traced
;
1242 pid_t pid
= task_pid_vnr(p
);
1244 if (unlikely(noreap
)) {
1246 int exit_code
= p
->exit_code
;
1250 read_unlock(&tasklist_lock
);
1251 if ((exit_code
& 0x7f) == 0) {
1253 status
= exit_code
>> 8;
1255 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1256 status
= exit_code
& 0x7f;
1258 return wait_noreap_copyout(p
, pid
, uid
, why
,
1263 * Try to move the task's state to DEAD
1264 * only one thread is allowed to do this:
1266 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1267 if (state
!= EXIT_ZOMBIE
) {
1268 BUG_ON(state
!= EXIT_DEAD
);
1272 traced
= ptrace_reparented(p
);
1274 if (likely(!traced
)) {
1275 struct signal_struct
*psig
;
1276 struct signal_struct
*sig
;
1279 * The resource counters for the group leader are in its
1280 * own task_struct. Those for dead threads in the group
1281 * are in its signal_struct, as are those for the child
1282 * processes it has previously reaped. All these
1283 * accumulate in the parent's signal_struct c* fields.
1285 * We don't bother to take a lock here to protect these
1286 * p->signal fields, because they are only touched by
1287 * __exit_signal, which runs with tasklist_lock
1288 * write-locked anyway, and so is excluded here. We do
1289 * need to protect the access to p->parent->signal fields,
1290 * as other threads in the parent group can be right
1291 * here reaping other children at the same time.
1293 spin_lock_irq(&p
->parent
->sighand
->siglock
);
1294 psig
= p
->parent
->signal
;
1297 cputime_add(psig
->cutime
,
1298 cputime_add(p
->utime
,
1299 cputime_add(sig
->utime
,
1302 cputime_add(psig
->cstime
,
1303 cputime_add(p
->stime
,
1304 cputime_add(sig
->stime
,
1307 cputime_add(psig
->cgtime
,
1308 cputime_add(p
->gtime
,
1309 cputime_add(sig
->gtime
,
1312 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1314 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1316 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1318 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1320 task_io_get_inblock(p
) +
1321 sig
->inblock
+ sig
->cinblock
;
1323 task_io_get_oublock(p
) +
1324 sig
->oublock
+ sig
->coublock
;
1325 spin_unlock_irq(&p
->parent
->sighand
->siglock
);
1329 * Now we are sure this task is interesting, and no other
1330 * thread can reap it because we set its state to EXIT_DEAD.
1332 read_unlock(&tasklist_lock
);
1334 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1335 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1336 ? p
->signal
->group_exit_code
: p
->exit_code
;
1337 if (!retval
&& stat_addr
)
1338 retval
= put_user(status
, stat_addr
);
1339 if (!retval
&& infop
)
1340 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1341 if (!retval
&& infop
)
1342 retval
= put_user(0, &infop
->si_errno
);
1343 if (!retval
&& infop
) {
1346 if ((status
& 0x7f) == 0) {
1350 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1353 retval
= put_user((short)why
, &infop
->si_code
);
1355 retval
= put_user(status
, &infop
->si_status
);
1357 if (!retval
&& infop
)
1358 retval
= put_user(pid
, &infop
->si_pid
);
1359 if (!retval
&& infop
)
1360 retval
= put_user(p
->uid
, &infop
->si_uid
);
1365 write_lock_irq(&tasklist_lock
);
1366 /* We dropped tasklist, ptracer could die and untrace */
1369 * If this is not a detached task, notify the parent.
1370 * If it's still not detached after that, don't release
1373 if (!task_detached(p
)) {
1374 do_notify_parent(p
, p
->exit_signal
);
1375 if (!task_detached(p
)) {
1376 p
->exit_state
= EXIT_ZOMBIE
;
1380 write_unlock_irq(&tasklist_lock
);
1389 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1390 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1391 * the lock and this task is uninteresting. If we return nonzero, we have
1392 * released the lock and the system call should return.
1394 static int wait_task_stopped(struct task_struct
*p
,
1395 int noreap
, struct siginfo __user
*infop
,
1396 int __user
*stat_addr
, struct rusage __user
*ru
)
1398 int retval
, exit_code
, why
;
1399 uid_t uid
= 0; /* unneeded, required by compiler */
1403 spin_lock_irq(&p
->sighand
->siglock
);
1405 if (unlikely(!task_is_stopped_or_traced(p
)))
1408 if (!(p
->ptrace
& PT_PTRACED
) && p
->signal
->group_stop_count
> 0)
1410 * A group stop is in progress and this is the group leader.
1411 * We won't report until all threads have stopped.
1415 exit_code
= p
->exit_code
;
1424 spin_unlock_irq(&p
->sighand
->siglock
);
1429 * Now we are pretty sure this task is interesting.
1430 * Make sure it doesn't get reaped out from under us while we
1431 * give up the lock and then examine it below. We don't want to
1432 * keep holding onto the tasklist_lock while we call getrusage and
1433 * possibly take page faults for user memory.
1436 pid
= task_pid_vnr(p
);
1437 why
= (p
->ptrace
& PT_PTRACED
) ? CLD_TRAPPED
: CLD_STOPPED
;
1438 read_unlock(&tasklist_lock
);
1440 if (unlikely(noreap
))
1441 return wait_noreap_copyout(p
, pid
, uid
,
1445 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1446 if (!retval
&& stat_addr
)
1447 retval
= put_user((exit_code
<< 8) | 0x7f, stat_addr
);
1448 if (!retval
&& infop
)
1449 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1450 if (!retval
&& infop
)
1451 retval
= put_user(0, &infop
->si_errno
);
1452 if (!retval
&& infop
)
1453 retval
= put_user((short)why
, &infop
->si_code
);
1454 if (!retval
&& infop
)
1455 retval
= put_user(exit_code
, &infop
->si_status
);
1456 if (!retval
&& infop
)
1457 retval
= put_user(pid
, &infop
->si_pid
);
1458 if (!retval
&& infop
)
1459 retval
= put_user(uid
, &infop
->si_uid
);
1469 * Handle do_wait work for one task in a live, non-stopped state.
1470 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1471 * the lock and this task is uninteresting. If we return nonzero, we have
1472 * released the lock and the system call should return.
1474 static int wait_task_continued(struct task_struct
*p
, int noreap
,
1475 struct siginfo __user
*infop
,
1476 int __user
*stat_addr
, struct rusage __user
*ru
)
1482 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1485 spin_lock_irq(&p
->sighand
->siglock
);
1486 /* Re-check with the lock held. */
1487 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1488 spin_unlock_irq(&p
->sighand
->siglock
);
1492 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1493 spin_unlock_irq(&p
->sighand
->siglock
);
1495 pid
= task_pid_vnr(p
);
1498 read_unlock(&tasklist_lock
);
1501 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1503 if (!retval
&& stat_addr
)
1504 retval
= put_user(0xffff, stat_addr
);
1508 retval
= wait_noreap_copyout(p
, pid
, uid
,
1509 CLD_CONTINUED
, SIGCONT
,
1511 BUG_ON(retval
== 0);
1517 static long do_wait(enum pid_type type
, struct pid
*pid
, int options
,
1518 struct siginfo __user
*infop
, int __user
*stat_addr
,
1519 struct rusage __user
*ru
)
1521 DECLARE_WAITQUEUE(wait
, current
);
1522 struct task_struct
*tsk
;
1525 add_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1527 /* If there is nothing that can match our critier just get out */
1529 if ((type
< PIDTYPE_MAX
) && (!pid
|| hlist_empty(&pid
->tasks
[type
])))
1533 * We will set this flag if we see any child that might later
1534 * match our criteria, even if we are not able to reap it yet.
1537 current
->state
= TASK_INTERRUPTIBLE
;
1538 read_lock(&tasklist_lock
);
1541 struct task_struct
*p
;
1543 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1544 int ret
= eligible_child(type
, pid
, options
, p
);
1548 if (unlikely(ret
< 0)) {
1550 } else if (task_is_stopped_or_traced(p
)) {
1552 * It's stopped now, so it might later
1553 * continue, exit, or stop again.
1556 if (!(p
->ptrace
& PT_PTRACED
) &&
1557 !(options
& WUNTRACED
))
1560 retval
= wait_task_stopped(p
,
1561 (options
& WNOWAIT
), infop
,
1563 } else if (p
->exit_state
== EXIT_ZOMBIE
&&
1564 !delay_group_leader(p
)) {
1566 * We don't reap group leaders with subthreads.
1568 if (!likely(options
& WEXITED
))
1570 retval
= wait_task_zombie(p
,
1571 (options
& WNOWAIT
), infop
,
1573 } else if (p
->exit_state
!= EXIT_DEAD
) {
1575 * It's running now, so it might later
1576 * exit, stop, or stop and then continue.
1579 if (!unlikely(options
& WCONTINUED
))
1581 retval
= wait_task_continued(p
,
1582 (options
& WNOWAIT
), infop
,
1585 if (retval
!= 0) /* tasklist_lock released */
1589 list_for_each_entry(p
, &tsk
->ptrace_children
,
1591 flag
= eligible_child(type
, pid
, options
, p
);
1594 if (likely(flag
> 0))
1600 if (options
& __WNOTHREAD
)
1602 tsk
= next_thread(tsk
);
1603 BUG_ON(tsk
->signal
!= current
->signal
);
1604 } while (tsk
!= current
);
1605 read_unlock(&tasklist_lock
);
1608 if (options
& WNOHANG
)
1610 retval
= -ERESTARTSYS
;
1611 if (signal_pending(current
))
1618 current
->state
= TASK_RUNNING
;
1619 remove_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1625 * For a WNOHANG return, clear out all the fields
1626 * we would set so the user can easily tell the
1630 retval
= put_user(0, &infop
->si_signo
);
1632 retval
= put_user(0, &infop
->si_errno
);
1634 retval
= put_user(0, &infop
->si_code
);
1636 retval
= put_user(0, &infop
->si_pid
);
1638 retval
= put_user(0, &infop
->si_uid
);
1640 retval
= put_user(0, &infop
->si_status
);
1646 asmlinkage
long sys_waitid(int which
, pid_t upid
,
1647 struct siginfo __user
*infop
, int options
,
1648 struct rusage __user
*ru
)
1650 struct pid
*pid
= NULL
;
1654 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1656 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1669 type
= PIDTYPE_PGID
;
1677 if (type
< PIDTYPE_MAX
)
1678 pid
= find_get_pid(upid
);
1679 ret
= do_wait(type
, pid
, options
, infop
, NULL
, ru
);
1682 /* avoid REGPARM breakage on x86: */
1683 asmlinkage_protect(5, ret
, which
, upid
, infop
, options
, ru
);
1687 asmlinkage
long sys_wait4(pid_t upid
, int __user
*stat_addr
,
1688 int options
, struct rusage __user
*ru
)
1690 struct pid
*pid
= NULL
;
1694 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1695 __WNOTHREAD
|__WCLONE
|__WALL
))
1700 else if (upid
< 0) {
1701 type
= PIDTYPE_PGID
;
1702 pid
= find_get_pid(-upid
);
1703 } else if (upid
== 0) {
1704 type
= PIDTYPE_PGID
;
1705 pid
= get_pid(task_pgrp(current
));
1706 } else /* upid > 0 */ {
1708 pid
= find_get_pid(upid
);
1711 ret
= do_wait(type
, pid
, options
| WEXITED
, NULL
, stat_addr
, ru
);
1714 /* avoid REGPARM breakage on x86: */
1715 asmlinkage_protect(4, ret
, upid
, stat_addr
, options
, ru
);
1719 #ifdef __ARCH_WANT_SYS_WAITPID
1722 * sys_waitpid() remains for compatibility. waitpid() should be
1723 * implemented by calling sys_wait4() from libc.a.
1725 asmlinkage
long sys_waitpid(pid_t pid
, int __user
*stat_addr
, int options
)
1727 return sys_wait4(pid
, stat_addr
, options
, NULL
);